Question

Let’s suppose I have a following table:

UID int (11) not null,  FOREIGN KEY (UID) REFERENCES users(ID),
OID int (11),  FOREIGN KEY (OID) REFERENCES orders(ID),
primary key(UID,OID)

Note the primary key.

And, I wanted to allow to have one NULL in order ID (OID) for each user ID (UID).

Though after setting the primary key it automatically makes my OID field NOT NULL.

As a work around I’m thinking to store 0 instead to mark it as an unrelated field.

So, now the question, is it possible to allow a foreign key constraint to accept also zeros (0) apart from the references of the specified field?

Answer 1

UNIQUE KEY constraint can be used in this case:

CREATE TABLE users (id int not null, primary key (id));
INSERT INTO users VALUES (1), (2), (3);

CREATE TABLE orders (id int not null, primary key (id));
INSERT INTO orders VALUES (1), (2), (3);

CREATE TABLE users_orders (uid int NOT NULL, oid int DEFAULT NULL,
                 FOREIGN KEY (uid) REFERENCES users (id),
                 FOREIGN KEY (oid) REFERENCES orders (id),
                 UNIQUE KEY (uid, oid));

INSERT INTO users_orders (uid) VALUES (1);

Here’s a demo to play with.

But be aware that you (probably) will still be able to insert more than one combination of specific UID - NULL in your table ( I assume here that you use InnoDB ). For example, this query…

INSERT INTO users_orders o (uid) VALUES (1), (1);

… will succeed as well (as it won’t break any existing constraints).

If this should be strictly prohibited, you better use some dummy value both in users and orders table as well – to make the foreign key constraint in your intersection table usable.

As a sidenote, I admit I’m a bit surprised by this situation. Intersection tables are usually designed so that they don’t have any NULLable fields: it’s LEFT (or RIGHT) JOIN in query that should be used to include users that don’t have any orders, for example. Could you explain a bit why this decision (to make OID nullable) was chosen at first hand?